Cutting process parameter design method, device and equipment and storage medium

Abstract

The invention relates to the technical field of cutting process parameter design, in particular to a cutting process parameter design method, device and equipment and a storage medium, and the method comprises the steps: firstly obtaining a plurality of to-be-measured laser powers, calling a pre-trained target temperature prediction model to predict the plurality of to-be-measured laser powers, and obtaining a plurality of predicted temperatures; simulation operation is executed based on the multiple predicted temperatures to construct a first training sample set, an initial cutting parameter prediction model is constructed, the first training sample set is used for conducting multi-round iterative training on the initial cutting parameter prediction model to obtain a target cutting parameter prediction model, multiple to-be-measured temperatures are obtained, and multiple to-be-measured temperatures are obtained; and calling the target cutting parameter prediction model to predict each to-be-measured temperature to obtain a plurality of initial cutting process data sets, and screening in the plurality of initial cutting process data sets by adopting an SAC algorithm to obtain an optimized cutting process data set so as to realize efficient prediction and optimization of the critical cutting depth and the cutting force.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of cutting process parameter design, and in particular to a cutting process parameter design method, device, equipment and storage medium. BACKGROUND

[0002] Monocrystalline silicon is a key material for micro-nozzle plates, precision alignment components and optical detection windows of inkjet printing OLED (organic light-emitting diode) manufacturing, and has high hardness, wide band gap and excellent optical performance, but its intrinsic brittleness easily causes brittle fracture and subsurface damage in traditional ultra-precision machining, which seriously affects the service life and printing precision of the components. In order to realize plastic domain machining with high surface quality and shape precision, laser-assisted machining and single-point diamond turning technology are introduced into this field, which reduces the brittleness of silicon material through local thermal softening of laser, and combines nanoscale precision machining to inhibit crack propagation and improve critical cutting depth, meeting the reliability requirements of silicon-based components in high-frequency and high-precision environments for printing equipment.

[0003] However, the prior art still has many deficiencies. For example, the influence of temperature on the brittle-plastic transition mechanism is not systematically quantified, the nanoscale machining mechanism is difficult to observe directly through experiments, and the temperature measurement in experiments is easily disturbed by process parameters, the temperature field distribution error is large, and the nonlinear relationship between laser power and temperature is not fully modeled. In terms of numerical simulation, the traditional finite element method is difficult to simulate the large deformation and fracture process of nanoscale cutting due to grid distortion, and the smoothed particle hydrodynamics method can solve the grid distortion problem, but it does not systematically analyze the influence of temperature on the brittle-plastic transition, and cannot provide sufficient support for process optimization. The existing related patents rely on high-cost external monitoring equipment, or use simulation methods with limitations, or rely on experimental trial-and-error optimization of parameters, which not only has limited applicability, but also has problems such as complex optimization process and low efficiency, and is difficult to meet the high-precision machining requirements of key components of inkjet printing equipment. SUMMARY

[0004] In order to overcome the deficiencies of the prior art, the purpose of the present application is to provide a cutting process parameter design method, device, equipment and storage medium, which aims to realize efficient prediction and optimization of critical cutting depth and cutting force, and improve the critical cutting depth and surface quality of monocrystalline silicon components under laser-assisted conditions.

[0005] The first aspect of the present application provides a cutting process parameter design method, comprising: obtaining a plurality of to-be-tested laser powers, calling a pre-trained target temperature prediction model to respectively predict the plurality of to-be-tested laser powers, and obtaining a plurality of predicted temperatures; performing simulation operations based on the plurality of predicted temperatures to construct a first training sample set; constructing an initial cutting parameter prediction model, and performing multi-round iterative training on the initial cutting parameter prediction model by using the first training sample set to obtain a target cutting parameter prediction model; obtaining a plurality of to-be-tested temperatures, calling the target cutting parameter prediction model to respectively predict each to-be-tested temperature, and obtaining a plurality of initial cutting process data groups, each initial cutting process data group comprising the to-be-tested temperature and a cutting process parameter corresponding to the to-be-tested temperature; and performing screening in the plurality of initial cutting process data groups by using a SAC (Soft Actor-Critic) algorithm to obtain an optimized cutting process data group.

[0006] Optionally, in the first implementation manner of the first aspect of the present application, before the obtaining a plurality of to-be-tested laser powers, calling a pre-trained target temperature prediction model to respectively predict the plurality of to-be-tested laser powers, and obtaining a plurality of predicted temperatures, the method further comprises: constructing an initial simulation model; obtaining a plurality of laser power samples, performing simulation operations on each laser power sample based on the initial simulation model to obtain simulation results, and optimizing the initial simulation model based on the simulation results to obtain a target simulation model; obtaining a plurality of to-be-simulated laser powers, performing simulation operations on each to-be-simulated laser power based on the target simulation model to construct a second training sample set; constructing an initial temperature prediction model, and performing multi-round iterative training on the initial temperature prediction model based on the second training sample set to obtain a target temperature prediction model.

[0007] Optionally, in the second implementation manner of the first aspect of the present application, the performing simulation operations on each laser power sample based on the initial simulation model to obtain simulation results, and optimizing the initial simulation model based on the simulation results to obtain a target simulation model comprises: presetting a plurality of initial correction coefficient candidate values based on the initial simulation model; for each initial correction coefficient candidate value, calling the initial simulation model to respectively perform simulation operations on each laser power sample to obtain a simulation feature width data corresponding to each laser power sample; obtaining an experimental feature width data corresponding to each laser power sample; and optimizing the initial simulation model based on the plurality of initial correction coefficient candidate values, the plurality of simulation feature width data, and the plurality of experimental feature width data to obtain a target simulation model.

[0008] Optionally, in a third implementation form of the first aspect of the present application, the optimizing the initial simulation model based on the plurality of initial correction coefficient candidate values, the plurality of simulation characteristic width data and the plurality of experimental characteristic width data to obtain a target simulation model comprises: calculating, for each of the initial correction coefficient candidate values, a relative error between the simulation characteristic width data and the experimental characteristic width data corresponding to each of the laser power samples to obtain a plurality of relative error rates; obtaining a preset relative error threshold, and comparing each of the relative error rates with the relative error threshold; when the relative error rates of all the laser power samples under the initial correction coefficient candidate value are less than or equal to the relative error threshold, taking the initial correction coefficient candidate value as a target correction coefficient value; and optimizing the initial simulation model based on the target correction coefficient value to obtain the target simulation model.

[0009] Optionally, in a fourth implementation form of the first aspect of the present application, the performing simulation operations based on the plurality of predicted temperatures to construct a first training sample set comprises: constructing an SPH simulation (Smoothed Particle Hydrodynamics) model for simulation; performing simulation operations based on the SPH simulation model for each of the predicted temperatures to obtain cutting force data corresponding to each of the predicted temperatures; constructing an initial relationship curve based on the plurality of predicted temperatures and the plurality of cutting force data; performing smoothing and denoising processing on the initial relationship curve by using a Gaussian convolution algorithm to obtain a target relationship curve; obtaining a brittle-plastic transition depth corresponding to each of the predicted temperatures based on the target relationship curve; and constructing the first training sample set based on the plurality of predicted temperatures, the plurality of cutting force data and the plurality of brittle-plastic transition depths.

[0010] Optionally, in a fifth implementation form of the first aspect of the present application, the screening, by using the SAC algorithm, from the plurality of initial cutting process data sets to obtain an optimized cutting process data set comprises: constructing an initial intelligent agent by using the SAC algorithm; obtaining a predefined reward function, and performing multi-round iteration update on network parameters of the initial intelligent agent based on the reward function and the plurality of initial cutting process data sets; when a preset iteration stop condition is met, taking the updated intelligent agent as a target intelligent agent; and performing screening operation on the plurality of initial cutting process data sets based on the target intelligent agent to obtain the optimized cutting process data set.

[0011] Optionally, in a sixth implementation form of the first aspect of the application, before the obtaining the predefined reward function, the method further comprises: determining a plurality of initial index parameters based on the initial cutting process data set; performing dimension unification processing on the plurality of index parameters respectively by using a normalization algorithm to obtain a plurality of normalized index parameters; obtaining a preset process priority requirement rule, and performing weight distribution on each of the normalized index parameters based on the process priority requirement rule to obtain a weight coefficient corresponding to each of the normalized index parameters; and defining the reward function based on the plurality of normalized index parameters and the plurality of weight coefficients.

[0012] The second aspect of the application provides a cutting process parameter design device, comprising: a temperature prediction module configured to obtain a plurality of to-be-tested laser powers, call a pre-trained target temperature prediction model to predict the plurality of to-be-tested laser powers respectively, and obtain a plurality of predicted temperatures; a simulation module configured to perform a simulation operation based on the plurality of predicted temperatures to construct a first training sample set; a training module configured to construct an initial cutting parameter prediction model, perform multi-round iterative training on the initial cutting parameter prediction model by using the first training sample set to obtain a target cutting parameter prediction model; a cutting parameter prediction module configured to obtain a plurality of to-be-tested temperatures, call the target cutting parameter prediction model to predict each of the to-be-tested temperatures respectively, and obtain a plurality of initial cutting process data sets, each of the initial cutting process data sets comprising the to-be-tested temperature and a cutting process parameter corresponding to the to-be-tested temperature; and an optimization module configured to perform screening in the plurality of initial cutting process data sets by using a SAC algorithm to obtain an optimized cutting process data set.

[0013] The third aspect of the application provides a cutting process parameter design device, comprising: a memory and at least one processor, the memory storing instructions; and at least one processor calling the instructions in the memory to enable the cutting process parameter design device to perform each step of the cutting process parameter design method described in any of the above aspects.

[0014] The fourth aspect of the application provides a computer-readable storage medium, the computer-readable storage medium storing instructions, the instructions being executed by a processor to implement each step of the cutting process parameter design method described in any of the above aspects.

[0015] In the technical solution of the present application, firstly, a plurality of to-be-tested laser powers are acquired, a pre-trained target temperature prediction model is called to predict the plurality of to-be-tested laser powers respectively, a plurality of predicted temperatures are obtained, simulation operations are performed based on the plurality of predicted temperatures to construct a first training sample set, then an initial cutting parameter prediction model is constructed, the first training sample set is used to perform multi-round iterative training on the initial cutting parameter prediction model to obtain a target cutting parameter prediction model, then a plurality of to-be-tested temperatures are acquired, the target cutting parameter prediction model is called to predict each to-be-tested temperature respectively, a plurality of initial cutting process data groups are obtained, each cutting process data group includes a to-be-tested temperature and a cutting process parameter corresponding to the to-be-tested temperature, finally, a SAC algorithm is used to screen the plurality of initial cutting process data groups to obtain an optimized cutting process data group, aiming to realize efficient prediction and optimization of the critical cutting depth and the cutting force, and improve the critical cutting depth and the surface quality of a monocrystalline silicon element under laser-assisted conditions, the method effectively overcomes the problems of brittle fracture and subsurface damage, enhances the processing stability, and can meet the reliability and service life requirements of key silicon-based parts of an inkjet printing device in a high-frequency and high-precision environment. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 A logical flowchart of a cutting process parameter design method provided for an embodiment of the present application is shown in the figure. Figure 2 A relationship curve diagram of a relative error rate and a correction coefficient provided for an embodiment of the present application is shown in the figure. Figure 3 A structural schematic diagram of a cutting process parameter design device provided for an embodiment of the present application is shown in the figure. Figure 4 A structural schematic diagram of a cutting process parameter design device provided for an embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0017] The present application provides a cutting process parameter design method, device, equipment and storage medium, in the present application, the terms "first", "second", "third", "fourth" and the like (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not have to be used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" or "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0018] For ease of understanding, the specific process of the embodiments of the present application is described below. Please refer to Figure 1 One embodiment of the cutting process parameter design method in the embodiments of the present application includes: 101, obtain a plurality of laser power to be measured, call a pre-trained target temperature prediction model to predict a plurality of the laser power to be measured respectively, and obtain a plurality of predicted temperatures; In this embodiment, the pre-trained target temperature prediction model is constructed based on a large amount of laser-assisted machining experimental data and high-precision temperature field simulation data through multiple rounds of algorithm iteration and parameter optimization. The nonlinear mapping relationship between laser power and temperature and the influence of processing environmental interference factors have been fully considered in the model training process, and the model has excellent generalization ability and prediction accuracy. In the prediction execution stage, each laser power to be measured is input into the target temperature prediction model in turn. The model quickly outputs the predicted temperature corresponding to each power value through the built-in feature extraction and mapping mechanism, realizing accurate quantitative conversion of laser power to processing temperature. The problems of large error, easy interference and high equipment cost caused by relying on thermometers, infrared detection and other means in traditional experimental measurement are effectively avoided, and the accuracy and reliability of temperature data acquisition are significantly improved.

[0019] 102, perform simulation operation based on a plurality of the predicted temperatures to construct a first training sample set; In this embodiment, based on the simulation operation performed on multiple predicted temperatures, the simulation operation is supported by the smoothed particle dynamics method as the core technology, relying on the LS-Dyna explicit dynamics simulation platform to build a three-dimensional numerical model suitable for the single crystal silicon ultra-precision beveling scene, giving full play to the advantages of the meshless characteristics of the method in simulating the nanoscale cutting large deformation and fracture process, and avoiding the grid distortion limitation of the traditional finite element method. In the simulation process, strict compliance with the established process parameter specification, including geometric simplification design of the local area of the workpiece (such as intercepting the local area of the workpiece to optimize the calculation efficiency), particle discretization processing of fixed particle size, tool modeling of specific blade radius and angle parameters, and boundary constraint setting of full freedom fixed on the bottom and left side, while controlling the tool movement according to the preset cutting speed and trajectory, ensuring that the maximum cutting depth covers the key interval of brittle-plastic transition. The predicted temperature as the core control variable is distributed in the simulation system according to the gradient of 20 to 800℃, forming multiple temperature working conditions, simulating the cutting response of single crystal silicon under different temperature environments, focusing on capturing key physical phenomena such as material plastic flow, crack initiation and propagation, stress distribution and hydrostatic pressure evolution, and synchronously recording the change data of cutting force with cutting depth, and through Gaussian convolution processing of the original cutting force signal, accurately positioning the critical value of brittle-plastic transition depth at each temperature. Based on each predicted temperature, the simulation output of brittle-plastic transition depth and cutting force and other key process and material response indicators, according to the unified data specification, the simulation data under multiple temperature domains and multiple working conditions are integrated and sorted, forming the first training sample set. With the simulation advantages of the SPH method, the precise reproduction of the nanoscale cutting micro-mechanism and brittle-plastic transition dynamic process that is difficult to observe in traditional experiments is realized, effectively making up for the limitations of entity experiments in microscale observation and multi-temperature domain parameter regulation. Through the systematic integration and accurate processing of simulation data, the first training sample set constructed has the characteristics of high data consistency and complete associated information, avoiding data fluctuations caused by environmental interference, equipment errors and other factors in entity experiments, providing high-quality and reliable data support for the training of the cutting force and brittle-plastic transition depth prediction model.

[0020] 103. constructing an initial cutting parameter prediction model, and performing multiple rounds of iterative training on the initial cutting parameter prediction model using the first training sample set to obtain a target cutting parameter prediction model; In this embodiment, when constructing the initial cutting parameter prediction model, a BP neural network is selected as the core architecture of the initial model, taking into account the nonlinear mapping characteristics of process parameters and performance indicators in the ultra-precision machining of single-crystal silicon. The first training sample set integrates the temperature-brittle-plastic transition depth data obtained from SPH cutting simulation and the measured results verified by the oblique cutting experiment. Among them, the measured results verified by the oblique cutting experiment refer to the processing and performance data obtained through actual operation in the dynamic load simulation experiment of continuous cutting depth of single-crystal silicon. This experiment uses a continuous cutting depth from 0 to 500 nm as a variable, and combines laser power to control the temperature of the workpiece deformation zone. It actually observes and records the state evolution of single-crystal silicon during the cutting process, including the actual position of the critical depth of brittle-plastic transition (i.e., the cutting depth threshold from plastic deformation dominance to brittle fracture dominance) under different temperature conditions, the processing state of the tool-workpiece contact area (such as the actual range of plastic flow, the specific process of crack initiation and propagation), and the quality characteristics of the processed surface (such as the presence and distribution of defects). It also includes actual measurement data of the cutting force changing with the cutting depth during the cutting process. The core function of these measured results is to verify and calibrate the simulation data obtained through SPH simulation. By comparing the deviation between the measured data and the simulation results, the parameter configuration of the cutting parameter prediction model is corrected, thereby improving the fit between the model prediction results and the actual processing conditions, and ensuring that the model can accurately reflect the real process response in the ultra-precision machining of single crystal silicon.

[0021] Before training using the first training sample set, the sample data needs to be standardized and preprocessed to eliminate dimensional differences. Then, the samples are input into the initial BP neural network model (backpropagation neural network model) in batches. The loss function is calculated based on the deviation between the predicted value and the true label of the sample. Through multiple iterations, the weights and threshold parameters of the network are adjusted. Simultaneously, oblique experimental data is introduced to verify the model's prediction results after each iteration. The experimental deviation is used to correct the model parameters until the model's prediction accuracy meets the preset error threshold, ultimately obtaining the target cutting parameter prediction model. The trained target cutting parameter prediction model can directly predict laser power to cutting performance indicators, replacing the inefficient traditional process of sequentially conducting simulations and experiments, significantly shortening the parameter prediction cycle.

[0022] 104. Obtain multiple temperatures to be measured, and call the target cutting parameter prediction model to predict each of the temperatures to be measured to obtain multiple initial cutting process data sets. Each initial cutting process data set includes the temperature to be measured and the cutting process parameters corresponding to the temperature to be measured. In this embodiment, the acquisition of the to-be-tested temperature needs to be combined with the temperature response characteristics of the brittle-plastic transition in the ultra-precision machining of single crystal silicon, and a plurality of temperature values covering the key process interval are selected, usually covering the high temperature range suitable for normal temperature to plastic domain machining, and containing the characteristic temperatures corresponding to the brittle-plastic transition threshold and performance jump, so as to ensure the representativeness and process coverage of the to-be-tested temperature. When calling the target cutting parameter prediction model that has completed training and calibration, each to-be-tested temperature is input as an input feature into the model. The model can quickly output the cutting process parameters corresponding to the temperature, including the critical depth of brittle-plastic transition and cutting force, and other core parameters, based on the nonlinear mapping relationship between temperature and cutting process parameters formed in the previous training. Then, each to-be-tested temperature and the corresponding cutting process parameters are integrated to form an initial cutting process data set.

[0023] 105. The SAC algorithm is used to screen the plurality of initial cutting process data sets to obtain an optimized cutting process data set.

[0024] In this embodiment, when the SAC algorithm is used to screen the initial cutting process data set, the characteristics of the algorithm suitable for continuous parameter optimization are relied on, and the entropy maximization exploration strategy is combined with the state input of each to-be-tested temperature and the corresponding cutting process parameters in the data set. While traversing the typical working conditions of the existing data set, the exploration of potential high-quality process working conditions is expanded, so as to avoid the selection process from falling into local optimization and missing the globally better process combination. The reward value of each data set is calculated based on the normalized brittle-plastic transition depth, cutting force, and temperature as the core dimensions, and the decision strategy of the Actor-Critic architecture collaborative optimization algorithm is used. The strategy continuously adjusts the selection tendency of the initial data set, gradually focuses on the candidate data set with better reward value, and the evaluation accuracy of the Critic module for the process value of the data set is continuously improved. This iterative process will continue until the decision strategy of the algorithm converges, that is, the fluctuation amplitude of the reward value is stabilized within the preset threshold. At this time, the process data set matched with the converged decision strategy is selected from the plurality of initial cutting process data sets, which is the optimized cutting process data set. The convergence mechanism of the decision strategy of the SAC algorithm effectively avoids the limitation of local optimization, and can accurately locate the globally optimal process combination that takes into account the improvement of the brittle-plastic transition depth, the reduction of the cutting force, and the reasonable control of the temperature. At the same time, the iterative optimization of the decision strategy does not need to rely on repeated trial and error of entity experiments, which greatly shortens the period of data set screening. The optimized cutting process data set obtained ultimately can directly match the requirements of high surface quality, low energy consumption, and low tool wear in the ultra-precision machining of single crystal silicon, and improves the stability and process economy of the machining process.

[0025] Before the obtaining of the plurality of predicted temperatures by calling the pre-trained target temperature prediction model to respectively predict the plurality of laser power to be measured, the embodiment of the present application further comprises: constructing an initial simulation model; obtaining a plurality of laser power samples, respectively performing simulation operation on each of the laser power samples based on the initial simulation model to obtain simulation results, and optimizing the initial simulation model based on the simulation results to obtain a target simulation model; obtaining a plurality of laser power to be simulated, performing simulation operation on each of the laser power to be simulated based on the target simulation model to construct a second training sample set; constructing an initial temperature prediction model, and performing multi-round iterative training on the initial temperature prediction model based on the second training sample set to obtain a target temperature prediction model.

[0026] In the embodiment, the construction of the initial simulation model needs to rely on the COMSOL multi-physics simulation platform to build a thermal conduction model of single crystal silicon laser heating, and the model simplification and boundary condition configuration are completed in combination with the process scene of single crystal silicon nanometric cutting. Specifically, the laser scanning is equivalent to a moving heat source and the scanning speed is set to 1000 mm / min, the influence of the diamond tool on the laser is simplified as energy attenuation by ignoring the physical modeling of the diamond tool, a 60 μm laser spot is adopted and the main irradiation area of the laser is subjected to grid refinement, and at the same time, the silicon wafer model is simplified to 3 × 3 × 1.5 mm to balance the calculation efficiency and simulation accuracy. After obtaining a plurality of laser power samples covering different energy intervals, the thermal conduction simulation is respectively performed on each sample based on the initial simulation model to obtain simulation results containing temperature field distribution, and the initial simulation model is optimized based on the simulation results to obtain a target simulation model. Then, a plurality of laser power to be simulated covering the adaptive energy range of single crystal silicon processing is obtained, and the parameterized simulation of the target simulation model is performed on each laser power to be simulated with a step of 0.5 W to obtain the stable temperature data of the workpiece surface corresponding to each laser power. At the same time, the data accuracy is improved through the interpolation method, and the mapping relationship between the laser power and the corresponding temperature is integrated into a structured data set, that is, a second training sample set is constructed, which breaks through the problems of insufficient spatial resolution and physical interference in the direct measurement of the temperature of the deformation zone in single crystal silicon nanometric cutting. Through the model calibration combining experiment and simulation, the temperature field prediction accuracy of the target simulation model is greatly improved, and the reliability and accuracy of the second training sample set are ensured.

[0027] In constructing the initial temperature prediction model, a neural network structure with one layer of 4 nodes is adopted, wherein the activation function of the hidden layer is selected as tansig (hyperbolic tangent activation function), and the activation function of the output layer is selected as purelin (linear transfer function); the initial model is trained by multiple rounds of iteration based on the second training sample set, with the laser power as the input and the corresponding temperature as the output, the deviation between the predicted value and the actual temperature of the sample is minimized by adjusting the network weight and threshold, until the prediction accuracy of the model meets the preset requirement, and finally the target temperature prediction model is obtained, realizing the rapid and accurate mapping from the laser power to the temperature, replacing the high-time-consuming process of frequent simulation, significantly reducing the acquisition cost of temperature data, and providing high-quality temperature basic data for the subsequent optimization of cutting process parameters.

[0028] In the embodiment of the application, the simulation operation is performed on each laser power sample based on the initial simulation model to obtain simulation results, and the initial simulation model is optimized based on the simulation results to obtain a target simulation model, including: presetting a plurality of initial correction coefficient candidate values based on the initial simulation model; for each initial correction coefficient candidate value, calling the initial simulation model to perform simulation operation on each laser power sample to obtain simulation feature width data corresponding to each laser power sample; obtaining experimental feature width data corresponding to each laser power sample; and optimizing the initial simulation model based on a plurality of initial correction coefficient candidate values, a plurality of simulation feature width data and a plurality of experimental feature width data to obtain a target simulation model.

[0029] In the embodiment, when the initial correction coefficient candidate values are preset based on the initial simulation model, the influence of energy attenuation in the laser transmission process on the single crystal silicon heating temperature field needs to be combined to select a candidate value combination covering a reasonable variation interval, which is usually set to a range of 0.6 to 1.1 and a step of 0.1 to fully cover the potential adjustment range of the heat source energy distribution. For each initial correction coefficient candidate value, it is substituted into the Gaussian heat source power density distribution formula of the initial simulation model, and then a heat conduction simulation operation is performed on each laser power sample to simulate the temperature field evolution process of the single crystal silicon surface under the corresponding power, and then the feature width data corresponding to the single crystal silicon boiling point isothermal line is extracted to obtain the simulation feature width data matched with each correction coefficient candidate value and each laser power sample. The experimental feature width data corresponding to each laser power sample needs to be obtained through a high-power ablation experiment: using the same power parameters as the laser power sample, laser scanning is completed on the single crystal silicon surface at a speed of 1000 mm / min, the surface morphology of the ablation zone is observed using a white light interferometer, the width of the completely ablated area (i.e. the feature area corresponding to the single crystal silicon boiling point isothermal line) is identified and measured, and then the experimental feature width data corresponding to each laser power sample is obtained. Then, the initial simulation model is optimized based on the plurality of simulation feature width data and the plurality of experimental feature width data to obtain the target simulation model. Through the iteration of multiple correction coefficient candidate values and the comparison with experimental data, the influence of laser transmission energy attenuation on the temperature field simulation result is effectively corrected, the output of the target simulation model is highly consistent with the actual processing thermal response characteristics, and the calibrated target simulation model can accurately reproduce the single crystal silicon temperature field distribution under different laser powers, providing high-precision simulation support for efficient acquisition of subsequent temperature data.

[0030] In the embodiment, the initial simulation model is optimized based on the plurality of initial correction coefficient candidate values, the plurality of simulation feature width data, and the plurality of experimental feature width data to obtain a target simulation model, which includes: for each initial correction coefficient candidate value, the relative error between the simulation feature width data and the experimental feature width data corresponding to each laser power sample is calculated to obtain a plurality of relative error rates; a preset relative error threshold is obtained, and each relative error rate is compared with the relative error threshold; when the relative error rates of all laser power samples under the initial correction coefficient candidate value are less than or equal to the relative error threshold, the initial correction coefficient candidate value is taken as a target correction coefficient value; and the initial simulation model is optimized based on the target correction coefficient value to obtain the target simulation model.

[0031] In the embodiment, for each initial correction coefficient candidate value, it needs to be substituted into the modified Gaussian heat source power density formula adopted by the initial simulation model. The modified Gaussian heat source power density formula is the core input model of the simulation of the laser heating temperature field of monocrystalline silicon, and the core parameter is the correction coefficient, which directly determines the distribution form of laser energy on the surface of monocrystalline silicon. Different correction coefficients will change the power density distribution of the heat source, and then affect the temperature field characteristics obtained by simulation, and finally reflect the difference in the simulation feature width data corresponding to the boiling point 3173℃ of monocrystalline silicon. Specifically, the calculation expression of the modified Gaussian heat source power density formula is: , wherein, represents the laser power density at a certain calculation point in space, that is, the laser energy input per unit area, and its value reflects the concentration of laser energy at this position, represents the correction coefficient, which is a dimensionless parameter, and is used to quantify the energy attenuation effect caused by the absorption and scattering of the tool during the laser passing through the diamond tool. When there is no energy attenuation, its value is defaulted to 1, represents the laser spot radius, which is a characteristic parameter for defining the spatial range of the Gaussian spot, and usually refers to the radius corresponding to the energy reduced to 1 / e² of the central peak, represents the spatial distance between the calculation point and the center of the laser light source, which is a parameter representing the position of the calculation point, represents the input laser power.

[0032] When optimizing the initial simulation model based on multiple simulation feature width data and experimental feature width data, the correction logic of the Gaussian heat source model needs to be combined. For each initial correction coefficient candidate value, the relative error corresponding to all laser power samples under the candidate value is calculated to obtain multiple relative error rates. The specific calculation expression is: , wherein, represents the relative error rate, represents the simulation feature width data, represents the experimental feature width data.

[0033] Subsequently, a preset relative error threshold is called, which is mainly set according to the experimental measurement accuracy, the actual adaptation ability of the simulation model, and the accuracy requirement of subsequent process application, and is generally set to 10%. The relative error rates of all laser power samples corresponding to each initial correction coefficient candidate value are compared with the threshold in turn. Only when the relative error rates of all laser power samples under a certain initial correction coefficient candidate value are less than or equal to the threshold, it is determined that the candidate value can realize effective matching between the simulation and experimental data in the full power interval, and then it is determined as the target correction coefficient value. For example, as Figure 2As shown in the figure, the horizontal axis represents the correction coefficient, the vertical axis represents the relative error rate corresponding to each laser power sample, and the three curves correspond to the relative error variation trend of the three groups of laser power samples 21.92W, 29.58W and 35.54W respectively. When the correction coefficient is 0.7, the relative errors of the simulation width and the measured width under the three powers are all within the 10% error band, and it is determined that the model correction is completed. The target correction coefficient value is substituted into the Gaussian heat source formula of the initial simulation model to complete the calibration and model optimization of the heat source parameters. The model adjusted in this way is the target simulation model. =0.7, the relative errors of the simulation width and the measured width under the three powers are all within the 10% error band, and it is determined that the model correction is completed. The target correction coefficient value is substituted into the Gaussian heat source formula of the initial simulation model to complete the calibration and model optimization of the heat source parameters. The model adjusted in this way is the target simulation model.

[0034] In the embodiment of the application, the simulation operation is performed based on the plurality of predicted temperatures to construct a first training sample set, comprising: constructing an SPH simulation model for simulation; performing simulation operation on each of the predicted temperatures based on the SPH simulation model to obtain cutting force data corresponding to each of the predicted temperatures; constructing an initial relationship curve based on the plurality of predicted temperatures and the plurality of cutting force data; performing smoothing and noise reduction processing on the initial relationship curve using a Gaussian convolution algorithm to obtain a target relationship curve; obtaining a brittle-plastic transition depth corresponding to each of the predicted temperatures based on the target relationship curve; and constructing the first training sample set based on the plurality of predicted temperatures, the plurality of cutting force data and the plurality of brittle-plastic transition depths.

[0035] In the embodiment, when performing simulation operation on each predicted temperature based on the SPH simulation model, the predicted temperature is substituted into the model as a variable, a uniform temperature field corresponding to the temperature condition is set, the cutting behavior of the tool advancing along the preset trajectory in the single crystal silicon beveling process is simulated, and the continuous data of the cutting force changing with the cutting depth are recorded synchronously, so that the cutting force data corresponding to each predicted temperature is obtained, which directly represents the dynamic evolution of the material removal mode at different temperatures. Then, an initial relationship curve is constructed based on the plurality of predicted temperatures and the corresponding cutting force data, with temperature as the horizontal axis and cutting force as the vertical axis, each predicted temperature and the corresponding cutting force key value are mapped as coordinate points in turn, and the correlation curve of temperature and cutting force is drawn, so that the evolution law of the cutting force in different temperature intervals is intuitively presented, and the negative regulation characteristics of temperature on the cutting force are clearly reflected. Considering that the original cutting force data is easily affected by factors such as particle dispersion error in the simulation process and local stress concentration caused by built-up edge, random noise and local outliers exist, the Gaussian convolution algorithm is used to perform smoothing and noise reduction processing on the initial relationship curve, the noise interference is weakened through convolution operation, the abnormal data points are corrected, the variation trend of the curve is more continuous and stable, and finally the target relationship curve is obtained.

[0036] In the embodiment, the evolution law of the cutting force with the depth of cut is as follows: in the initial stage of cutting, when the depth of cut is in a lower range, the single crystal silicon is mainly removed in a plastic flow mode, the workpiece particles form chips through continuous slip, the first deformation zone of the contact surface between the tool and the workpiece stably expands, at this time, the cutting force shows a smooth upward trend, the amplitude and frequency of the fluctuation are at a lower level, and the fitting deviation value after Gaussian convolution processing is also stable, reflecting the continuity and regularity of material deformation; as the depth of cut continuously increases, when the brittle-plastic transition critical point is reached, micro-cracks begin to emerge in the single crystal silicon, the material removal mode gradually transitions from plastic dominance to brittle dominance, and enters the initial state of brittle-plastic mixed phase change, a significant feature of this stage is that the amplitude and frequency of the cutting force fluctuation increase synchronously; the emergence and small expansion of cracks will cause instantaneous changes in cutting load, leading to an intensification of the fluctuation of the force signal, but since the cracks have not formed a continuous network of expansion at this time, the material still has partial plastic response, the fitting deviation value after Gaussian convolution still maintains at a low level and does not fluctuate greatly, accurately characterizing the transition characteristics of the mixed phase change; when the depth of cut further increases and exceeds the brittle-plastic transition critical range, the single crystal silicon enters the brittle-dominated period, at this time, the micro-cracks in the material rapidly interconnect and diffuse to form a crack network, the material removal is mainly in the form of discontinuous collapse and separation, the load transfer in the cutting process is sharply unstable, showing violent oscillation of the cutting force, and the Gaussian convolution fitting deviation also greatly increases, reflecting the dominant effect of brittle fracture on the cutting mechanical behavior. According to the above evolution law of the cutting force with the depth of cut, the starting position of the synchronous increase of the amplitude and frequency of the cutting force fluctuation in the initial state of the brittle-plastic mixed phase change is defined as the brittle-plastic transition depth of the single crystal silicon based on the target relationship curve, so as to obtain the brittle-plastic transition depth corresponding to each predicted temperature. Finally, based on multiple predicted temperatures, corresponding cutting force core data and accurately extracted brittle-plastic transition depths, a structured data set is integrated according to unified data specifications, that is, the first training sample set is constructed.

[0037] In the embodiment of the present application, the SAC algorithm is used to screen the plurality of initial cutting process data sets to obtain an optimized cutting process data set, which comprises: constructing an initial agent by using the SAC algorithm; obtaining a pre-defined reward function, and performing multiple rounds of iterative updates on the network parameters of the initial agent based on the reward function and the plurality of initial cutting process data sets; when a pre-set iteration stopping condition is met, the updated agent is taken as a target agent; and performing a screening operation on the plurality of initial cutting process data sets based on the target agent to obtain the optimized cutting process data set.

[0038] In this embodiment, when the SAC algorithm is used to construct the initial agent, the Actor-Critic architecture that meets the optimization requirements of the single crystal silicon cutting process parameters is built by fully relying on the adaptability of the algorithm to the continuous parameter optimization. Among them, the Actor network undertakes the output function of the temperature control action, and generates the temperature suggestion in the continuous interval according to the preset strategy; the Critic network is responsible for evaluating the comprehensive process value corresponding to the temperature, and outputs the quantitative Q value; at the same time, the entropy maximization strategy is integrated to ensure that the agent actively expands the unknown temperature field while exploring the known effective temperature interval, thereby avoiding the limitations of local optimal solution from the root and laying a foundation for mining the global optimal process parameters. Based on the predefined reward function and multiple initial cutting process data sets, the initial agent performs multiple rounds of iterative updates. Specifically, the agent selects the temperature parameter from the multiple initial cutting process data sets as the control action, and substitutes the brittle-plastic transition depth and cutting force in the initial cutting process data set into the reward function for calculation to obtain the reward value corresponding to the temperature. Then, the experience data such as the control action and the reward value are stored in the experience replay pool. The Actor network dynamically adjusts the strategy according to these experiences to optimize the temperature output tendency, and the Critic network continuously corrects the value evaluation model to improve the Q value prediction accuracy. The two work together to complete the iterative update of the network parameters. When the preset iteration stopping condition is met, that is, the reward value fluctuation amplitude is stable within the preset threshold, the reward mean of continuous multiple iterations has no significant improvement, or the preset iteration number is reached, the parameter update is stopped. At this time, the agent is the target agent.

[0039] Based on the target agent screening in multiple initial cutting process data sets, the Actor network of the target agent accurately identifies the optimal temperature parameter that can balance the optimization target from the data set according to the optimized strategy. The set composed of the brittle-plastic transition depth, cutting force and other process data corresponding to the temperature is the optimized cutting process data set. The optimization target refers to finding the optimal temperature in the process parameter optimization of single crystal silicon ultra-precision machining to maximize the brittle-plastic transition depth, minimize the cutting force and temperature, so that the three achieve the optimal collaborative state under the preset weight of the reward function.

[0040] In this embodiment, before the predefined reward function is obtained, it further includes: determining multiple initial index parameters based on the initial cutting process data set; using a normalization algorithm to respectively perform dimensionless processing on multiple index parameters to obtain multiple normalized index parameters; obtaining a preset process priority requirement rule, and respectively assigning weights to each normalized index parameter based on the process priority requirement rule to obtain a weight coefficient corresponding to each normalized index parameter; and defining the reward function based on multiple normalized index parameters and multiple weight coefficients.

[0041] In this embodiment, when determining multiple initial index parameters based on the initial cutting process data set, it is necessary to closely adhere to the core optimization requirements of single-crystal silicon ultra-precision machining and extract key parameters directly related to process performance from the data set, specifically the brittle-plastic transition depth, cutting force, and temperature. Given the fundamental differences in the physical dimensions of the brittle-plastic transition depth, cutting force, and temperature, direct numerical calculations would lead to an imbalance in the weights of the optimization objectives. Therefore, a normalization algorithm is needed to scale the original values ​​of the three types of parameters to the range of 0 to 1. By eliminating the influence of different physical dimensions, the three types of parameters have an equal basis for numerical comparison, preventing any one index from dominating the reward calculation due to unit differences, and ensuring the fairness of subsequent weight allocation. After obtaining the preset process priority requirement rules, weights are assigned to each normalized index parameter based on these rules. The process priority requirement rules are guiding principles for prioritizing the core objective of "finding the optimal temperature to maximize the brittle-plastic transition depth and minimize cutting force and temperature" in combination with the specific requirements of actual processing scenarios. According to the process priority requirement rules, weight coefficients are assigned to the normalized index parameters corresponding to the brittle-plastic transition depth, cutting force, and temperature, and the constraint condition that the sum of the three types of coefficients is 1 is strictly followed, so that the reward function can accurately match the scenario requirements and guide the agent to evolve towards the preset optimization direction.

[0042] Next, a reward function is defined based on multiple normalized index parameters and their corresponding weighting coefficients. The specific calculation expression of the reward function is as follows: , in, R(T) Represents the reward function, T Indicates temperature. Depth(T) Indicates the depth of the brittle-plastic transition. Force(T) Indicates cutting force. w 1 represents the first weighting coefficient corresponding to the depth of the brittle-plastic transition. w 2 represents the second weighting coefficient corresponding to the cutting force. w 3 indicates the third weighting coefficient corresponding to temperature. Indicates the depth of brittle-plastic transition Depth (T) The result after normalization Indicates the cutting force Force(T) The result after normalization Indicates temperature T The result after normalization. The final defined reward function can transform multi-dimensional optimization objectives into intuitive quantitative signals, providing a clear learning guide for the agent of the SAC algorithm. This ensures that the agent can accurately capture the optimization patterns of process parameters during the iteration process, and ultimately find the global optimal solution that meets the needs of the scenario, thereby improving the intelligent optimization level and practical application value of monocrystalline silicon processing technology.

[0043] The cutting process parameter design method in the embodiments of the present application is described above, and the cutting process parameter design device in the embodiments of the present application is described below. Please refer to Figure 3 The cutting process parameter design device in the embodiments of the present application includes one embodiment: The temperature prediction module 301 is configured to obtain a plurality of to-be-tested laser powers, call a pre-trained target temperature prediction model to predict the plurality of to-be-tested laser powers respectively, and obtain a plurality of predicted temperatures. The simulation module 302 is configured to perform a simulation operation based on the plurality of predicted temperatures to construct a first training sample set. The training module 303 is configured to construct an initial cutting parameter prediction model, perform multi-round iterative training on the initial cutting parameter prediction model by using the first training sample set, and obtain a target cutting parameter prediction model. The cutting parameter prediction module 304 is configured to obtain a plurality of to-be-tested temperatures, call the target cutting parameter prediction model to predict each to-be-tested temperature respectively, and obtain a plurality of initial cutting process data groups, each of which includes the to-be-tested temperature and a cutting process parameter corresponding to the to-be-tested temperature. The optimization module 305 is configured to perform screening in the plurality of initial cutting process data groups by using a SAC algorithm, and obtain an optimized cutting process data group.

[0044] Based on the same idea as the method in the above embodiments, the device provided by the present application can implement the method of the above embodiments.

[0045] The cutting process parameter design device in the embodiments of the present application is described above Figure 3 The cutting process parameter design device in the embodiments of the present application is described in detail from the perspective of modular functional entities, and the cutting process parameter design device in the embodiments of the present application is described in detail from the perspective of hardware processing.

[0046] Figure 4is a structural schematic diagram of a cutting process parameter design device provided by an embodiment of the present application. The cutting process parameter design device 400 can have great differences due to different configurations or performances, and can include one or more central processing units (CPUs) 410 (for example, one or more processors) and a memory 420, one or more storage media 430 (for example, one or more mass storage devices) storing application programs 433 or data 432. The memory 420 and the storage media 430 can be temporary storage or persistent storage. The programs stored in the storage media 430 can include one or more modules (not shown in the figure), and each module can include a series of instruction operations in the cutting process parameter design device 400. Further, the processor 410 can be configured to communicate with the storage media 430 and execute the series of instruction operations in the storage media 430 on the cutting process parameter design device 400 to implement the steps of the cutting process parameter design method provided by the above-mentioned method embodiments.

[0047] The cutting process parameter design device 400 can also include one or more power supplies 440, one or more wired or wireless network interfaces 450, one or more input / output interfaces 460, and / or one or more operating systems 431, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, and the like. Those skilled in the art can understand that the cutting process parameter design device 400 can also include other components, and the components shown in the figure are not exhaustive. Figure 4 The cutting process parameter design device structure shown does not constitute a limitation on the cutting process parameter design device, and can include more or fewer components than shown, or combine certain components, or different component arrangements.

[0048] The present application also provides a computer readable storage medium, which can be a non-volatile computer readable storage medium or a volatile computer readable storage medium. The computer readable storage medium stores instructions, and when the instructions are run on a computer, the computer executes the steps of the cutting process parameter design method.

[0049] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system or device, unit can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0050] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or say the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0051] Finally, it should be noted that: the above only for the preferred examples of the present application, and not for limiting the present application, although the present application is described in detail with reference to the foregoing embodiments, for those skilled in the art, it still can modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method of cutting process parameter design, characterized by, The method comprises the following steps: obtaining a plurality of to-be-tested laser powers, calling a pre-trained target temperature prediction model to predict the plurality of to-be-tested laser powers respectively, and obtaining a plurality of predicted temperatures; performing simulation operations based on the plurality of predicted temperatures to construct a first training sample set; constructing an initial cutting parameter prediction model, and performing multi-round iterative training on the initial cutting parameter prediction model by using the first training sample set to obtain a target cutting parameter prediction model; obtaining a plurality of to-be-tested temperatures, calling the target cutting parameter prediction model to predict each to-be-tested temperature respectively, and obtaining a plurality of initial cutting process data groups, each initial cutting process data group comprising the to-be-tested temperature and a cutting process parameter corresponding to the to-be-tested temperature; screening the plurality of initial cutting process data groups by using a SAC algorithm to obtain an optimized cutting process data group.

2. The cutting process parameter design method according to claim 1, characterized by, Before the step of obtaining a plurality of to-be-tested laser powers, calling a pre-trained target temperature prediction model to predict the plurality of to-be-tested laser powers respectively, and obtaining a plurality of predicted temperatures, the method further comprises the following steps: constructing an initial simulation model; obtaining a plurality of laser power samples, performing simulation operations on each laser power sample based on the initial simulation model to obtain simulation results, and optimizing the initial simulation model based on the simulation results to obtain a target simulation model; obtaining a plurality of to-be-simulated laser powers, and performing simulation operations on each to-be-simulated laser power based on the target simulation model to construct a second training sample set; constructing an initial temperature prediction model, and performing multi-round iterative training on the initial temperature prediction model based on the second training sample set to obtain a target temperature prediction model.

3. The cutting process parameter design method according to claim 2, wherein, The step of performing simulation operations on each laser power sample based on the initial simulation model to obtain simulation results, and optimizing the initial simulation model based on the simulation results to obtain a target simulation model comprises the following steps: presetting a plurality of initial correction coefficient candidate values based on the initial simulation model; for each initial correction coefficient candidate value, calling the initial simulation model to perform simulation operations on each laser power sample to obtain a simulation feature width data corresponding to each laser power sample; obtaining an experimental feature width data corresponding to each laser power sample; optimizing the initial simulation model based on the plurality of initial correction coefficient candidate values, the plurality of simulation feature width data and the plurality of experimental feature width data to obtain a target simulation model.

4. The cutting process parameter design method according to claim 3, wherein, The step of optimizing the initial simulation model based on the plurality of initial correction coefficient candidate values, the plurality of simulation feature width data and the plurality of experimental feature width data to obtain a target simulation model comprises the following steps: for each initial correction coefficient candidate value, calculating a relative error between the simulation feature width data corresponding to each laser power sample and the experimental feature width data respectively to obtain a plurality of relative error rates; obtaining a preset relative error threshold, and comparing each relative error rate with the relative error threshold respectively; When the relative error rates of all laser power samples under the initial correction coefficient candidate value are less than or equal to the relative error threshold, the initial correction coefficient candidate value is taken as a target correction coefficient value; optimizing the initial simulation model based on the target correction coefficient value to obtain the target simulation model.

5. The cutting process parameter design method of claim 1, wherein, The simulation operation is performed based on the plurality of predicted temperatures to construct a first training sample set, including: constructing an SPH simulation model for simulation; performing simulation operations based on the SPH simulation model for each of the plurality of predicted temperatures to obtain cutting force data corresponding to each of the plurality of predicted temperatures; constructing an initial relationship curve based on the plurality of predicted temperatures and the plurality of cutting force data; performing smoothing and noise reduction processing on the initial relationship curve using a Gaussian convolution algorithm to obtain a target relationship curve; obtaining brittle-plastic transition depths corresponding to each of the plurality of predicted temperatures based on the target relationship curve; constructing the first training sample set based on the plurality of predicted temperatures, the plurality of cutting force data, and the plurality of brittle-plastic transition depths.

6. The cutting process parameter design method of claim 1, wherein, The SAC algorithm is used to screen the plurality of initial cutting process data sets to obtain an optimized cutting process data set, including: constructing an initial intelligent agent using the SAC algorithm; obtaining a predefined reward function, and performing multiple rounds of iterative updates on network parameters of the initial intelligent agent based on the reward function and the plurality of initial cutting process data sets, and when a preset iteration stopping condition is met, updating the intelligent agent as a target intelligent agent; performing a screening operation on the plurality of initial cutting process data sets based on the target intelligent agent to obtain the optimized cutting process data set.

7. The cutting process parameter design method according to claim 6, wherein, Before obtaining the predefined reward function, it further includes: determining a plurality of initial index parameters based on the initial cutting process data set; performing dimensionless processing on the plurality of index parameters using a normalization algorithm to obtain a plurality of normalized index parameters; obtaining a preset process priority requirement rule, and performing weight distribution on each of the plurality of normalized index parameters based on the process priority requirement rule to obtain a weight coefficient corresponding to each of the plurality of normalized index parameters; defining the reward function based on the plurality of normalized index parameters and the plurality of weight coefficients.

8. A cutting process parameter design apparatus characterized by comprising: including: a temperature prediction module for obtaining a plurality of to-be-tested laser powers, calling a pre-trained target temperature prediction model to predict the plurality of to-be-tested laser powers to obtain a plurality of predicted temperatures; a simulation module for performing simulation operations based on the plurality of predicted temperatures to construct a first training sample set; a training module for constructing an initial cutting parameter prediction model, and performing multiple rounds of iterative training on the initial cutting parameter prediction model using the first training sample set to obtain a target cutting parameter prediction model; a cutting parameter prediction module for obtaining a plurality of to-be-tested temperatures, calling the target cutting parameter prediction model to predict each of the plurality of to-be-tested temperatures to obtain a plurality of initial cutting process data sets, each of the plurality of initial cutting process data sets including the to-be-tested temperature and a cutting process parameter corresponding to the to-be-tested temperature; An optimization module is configured to filter the plurality of initial cutting process data sets using a SAC algorithm to obtain an optimized cutting process data set.

9. A cutting process parameter design apparatus characterized by comprising: The cutting process parameter design device comprises a memory and at least one processor, and the memory stores instructions. The at least one processor invokes the instructions in the memory to enable the cutting process parameter design device to perform the steps of the cutting process parameter design method according to any one of claims 1-7.

10. A computer-readable storage medium having stored thereon instructions, the computer-readable storage medium comprising: The instructions are executed by the processor to implement the steps of the cutting process parameter design method according to any one of claims 1-7.

Images